DeepPySR Advances Symbolic Regression for Scientific Discovery

A new framework called DeepPySR has been introduced to enhance symbolic regression, a method for deriving interpretable mathematical formulas directly from data. Unlike "black-box" AI models, symbolic regression provides transparent, "glass-box" equations, which are crucial for fields like clinical medicine and social science where interpretability is paramount. DeepPySR tackles key limitations of traditional symbolic regression, including handling numerous input variables, selecting optimal formulas from a set of trade-offs, and managing messy data. It achieves this through a dynamic pruning mechanism that removes irrelevant features, an advanced Pareto selection method that balances accuracy and complexity, and a hierarchical architecture for building complex equations. Empirical tests on physics benchmarks and real-world biomedical and social science datasets demonstrate DeepPySR's superior performance compared to existing tools, yielding more accurate and interpretable models for tasks such as predicting body fat, heart disease risk, and student performance.